Prediction of Relative Density in L-PBF Process
Description
This study focuses on predicting densification in laser powder bed fusion (L-PBF), where the relative density of a part influences its mechanical properties but is difficult and expensive to measure directly. The researchers developed and optimized six machine learning models—KNN, AdaBoost-DT, and four multilayer perceptrons trained with different algorithms (SCG, LM, BR, RB)—using parameters like laser power, scanning speed, and hatch spacing. These models' outputs were combined into a single, highly accurate predictive system called a committee machine intelligent system (CMIS). Statistical and visual analyses confirmed its reliability and superior accuracy compared to previous methods. An outlier detection process identified a few data points outside the model’s valid prediction range, underscoring the model's robustness and limitations.